Devstral 2 123B
Devstral 2 123B is a 123 billion parameter model from Mistral AI optimized for coding and development tasks. Features advanced reasoning capabilities for software engineering workflows.
Added Dec 9, 2025
Model weightsPricing
Auto routing · per 1M tokens- Input
- $0.40
- Output
- $1.40
Specifications
- Context window
- 262.1K
- Max output
- 65.5K
- Parameters
- 123B
- Avg output (7d)
- 545 tokens
- Longer than 51% of models
Benchmarks
Benchmarks
Sourced from Artificial Analysis.
Intelligence Index
8.6
Coding Index
31.3
Agentic Index
2.3
Agentic work
AutomationBench-AA
Workflow automation with guardrail penalties
3.1%
Better than 22% of models compared
AutomationBench-AA Tasks Completed
Fully completed workflows without guardrail violations
0.0%
Better than 8% of models compared
AA-Briefcase v1.1
Agentic knowledge work (Elo)
482 Elo
Better than 21% of models compared
GDPval-AA v2.1
Economically valuable tasks (Elo)
544 Elo
Better than 26% of models compared
Document reasoning
GDP.pdf
Professional PDF reasoning: all-pass rate
2.4%
Better than 16% of models compared
AA-LCR v1.1
Long context reasoning with updated grading
32.3%
Better than 31% of models compared
Reasoning
HLE
Humanity's Last Exam
3.6%
Better than 6% of models compared
IFBench
Instruction-following benchmark
38.1%
Better than 34% of models compared
CritPt
Research-level physics reasoning
0.0%
Coding
Terminal-Bench v4.0
Practical coding and terminal tasks
0.0%
Better than 16% of models compared
SciCode
Python programming for scientific computing
32.8%
Better than 12% of models compared
LiveCodeBench
Contamination-free coding benchmark
44.8%
Better than 51% of models compared
Math
AIME 2025
American Invitational Mathematics Examination 2025
36.7%
Better than 37% of models compared
Knowledge
MMLU-Pro
Professional and academic subject knowledge
76.2%
Better than 53% of models compared
AA-Omniscience Accuracy
Proportion of correctly answered questions
20.8%
AA-Omniscience Hallucination Rate
Rate of incorrect answers among non-correct responses
85.3%
Legacy benchmarks
GPQA Diamond (legacy)
Graduate-level scientific reasoning
59.4%
Better than 34% of models compared
Terminal-Bench Hard (legacy)
Agentic coding and terminal use
18.9%
Better than 59% of models compared
T²-Bench Telecom (legacy)
Conversational AI agents in dual-control scenarios
24.9%
Better than 28% of models compared
AA-LCR (unversioned / legacy)
Long context reasoning evaluation
32.3%
Better than 31% of models compared
GDPval-AA (unversioned / legacy)
Economically valuable tasks
2.2%
Last updated Oct 22, 2026
Artificial AnalysisProviders
Auto routing is available for this model. Explicit provider selection is not available.
Loading provider options…
Reliability
Last 3 days